Seminar 2007 03 23 Image Visual Features Classification

From CISSTwiki

Computer Science Department Seminar
Exploring Image and Video by Classification and Clustering on Global and Local Visual Features

Date: Friday, March 23, 2007
Time: 1:00 - 3:30
Place: Shaffer 100

Speaker: Le Lu
Johns Hopkins University
Title: Exploring Image and Video by Classification and Clustering on Global and Local Visual Features

Abstract

We present results on four specific computer vision problems involving unsupervised visual data partitioning, discriminative multiple-class classification and online adaptive appearance learning.

Our solution apply statistical machine learning techniques over extracted local and global scale visual features.

First, we develop a new clustering algorithm to exploit temporal video structures into piecewise elements ( a.k.a. video shot segmentation) by combining central and subspace constraints for a unified solution. The proposed algorithm is also demonstrated its applicability to illumination-invariant face clustering.

Second, we detect and recognize the spatialtemporal video subvolumes as action units using a trained 3D-surface action model via multi-scale temporal searching. The dynamic 3D-surface based action model is built up as an empirical distribution over the basic static posture elements in the spirit of texton representation.

Third, we train a discriminative-probabilistic multi-modal density classifier to evaluate the responses of 20 semantic material classes from a large collection of challenging home photos. Then the task of learning photo categories is based on the global image features extracted from the material class-specific density response maps over spatial domain. We adopt the classifier combination technique of a set of random weak discriminators to handle the complex multi-modal photo-feature distributions in high dimensional parameter space.

Finally, we present a unified nonparametric approach for three applications: location based dynamic template video tracking in low to medium resolution, segmentation based object-level image matching across viewpoints, and binary foreground/background segmentation tracking.

Bio

Le Lu is a Ph.D. Candidate in Computer Science Department at Johns Hopkins University where he received his MSE degree in 2004. Before he came to Johns Hopkins, he spent three years in National Laboratory of Patter Recognition, Institute of Automation, Chinese Academy of Sciences, studying computational geometry and computer vision. He also received his B.E. in Precision Instrument (major) and Automatic Control (minor) from Beijing Polytechnic Univerisity, China, in 1996.

Since October 2005, Le Lu has been a Research Scientist in the Department of Integrated Data Systems, Siemens Corporate Research, Inc. Princeton, New Jersey. During the summer of 2004, he was a Research Intern with Interactive Visual Media Group, Microsoft Research, Redmond, WA. He also worked in Visual Computing Group, Microsoft Research, Beijing from Dec. 1999 to July 2001. From March to August in 1999, he was a Visiting Scholar in Department of Electronic Engineering, Chinese University of Hong Kong, Hong Kong.